Changes made to 3rd submission:¶

scaling the input images to the discriminator so that the input to discriminator is the same from the real images and the generator.
I tried to run the generator twice during training as adviced by the reviewer
I ran train method twice once with uniform noise sampling and the other with normal distribution for the z_input (as adviced by the ganhack link: https://github.com/soumith/ganhacks )
I plotted the D and G loss ( in some output I even plotted the lose per each iteration)
I tried to change the hyperparameters like: learning rate, batch_size, beta and alpha
I tried to use smoothening for the label. Actually i used as follow:

d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_real, labels=tf.ones_like(d_output_real)*(1-smooth)))
d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_fake, labels=tf.zeros_like(d_output_fake)))

so only the real loss was smoothed to about .9 instead of 1

then I used the following method to smoothe even the fake as:

d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_real, labels=tf.ones_like(d_output_real)(1-smooth)))
d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_fake, labels=tf.ones_like(d_output_fake)smooth))

Most importantly I added one layer to the discriminator and two layers to the generator.

I am not happy with the result although it ws improved, and this makes me realize how difficult the GAN network to train.¶

I had a problem of the speed of my device; so I moved to the udacity workspace but it sometimes disconnect in the middle of training ; then I moved to aws ec2 p2.xlarge instance machine but still it disconnect in the middle of the training and I tried to research how to make the connection alive via ssh_config.¶

Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

#the following line didn't work on aws ec2 machine instance
#data_dir = '/data'

data_dir = '~/data'
!pip install matplotlib==2.0.2
# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Found mnist Data
Found celeba Data

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x7fb26e943128>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x7fb26e872940>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.0.0
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    real_input = tf.placeholder(tf.float32, [None, image_width,image_height, image_channels], name='real_input')
    z_input = tf.placeholder(tf.float32,[None, z_dim], name='z_input')
    lr = tf.placeholder(tf.float32, name= 'lr')

    return real_input, z_input, lr


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

#set alpha 

alpha = .2

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the discriminator, tensor logits of the discriminator).

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param images: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    # TODO: Implement Function

    #print(images.shape)
    with tf.variable_scope('discriminator',reuse=reuse):
        x = tf.layers.conv2d(inputs=images, filters = 32 , kernel_size= 5, strides= 2 ,  padding='same',activation= None, kernel_initializer=tf.contrib.layers.xavier_initializer())
        relul = tf.maximum(x * alpha, x)
        
   
        #now shape is 14*14*32
        
        x2 = tf.layers.conv2d(relul, filters = 64, kernel_size= 5, strides=2, padding = 'same',activation=None, kernel_initializer=tf.contrib.layers.xavier_initializer())
        x2 = tf.layers.batch_normalization(x2, training= True)
        relul2 = tf.maximum(x2 * alpha, x2)
        #relul2 = tf.layers.dropout(relul2,rate=.8)
        
        # Now shape is 7*7*64

        x3 =  tf.layers.conv2d(relul2, filters = 128, kernel_size= 5, strides=1, padding = 'same',activation=None, kernel_initializer=tf.contrib.layers.xavier_initializer())
        x3 =  tf.layers.batch_normalization(x3, training= True)
        relul3 = tf.maximum(x3 * alpha, x3)
        #relul3 = tf.layers.dropout(relul3,rate=.8)
        #print(relul3.shape)
        
        x4 =  tf.layers.conv2d(relul3, filters = 256, kernel_size= 5, strides=2, padding = 'same',activation=None, kernel_initializer=tf.contrib.layers.xavier_initializer())
        x4 =  tf.layers.batch_normalization(x4, training= True)
        relul4 = tf.maximum(x4 * alpha, x4)
        #print('relul4  =',relul4.shape)
        

        
        flat = tf.reshape(relul4, (-1, 4*4*256))
        
        logits = tf.layers.dense(flat, 1)
        
        output = tf.sigmoid(logits)
        
        
        
    return output, logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variables in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    
   
    # TODO: Implement Function

    reuse = False if is_train else True
    with tf.variable_scope('generator', reuse = reuse):
        
        x1= tf.layers.dense(z,2*2*1024, kernel_initializer=tf.contrib.layers.xavier_initializer())
        #print(x1.shape)
        x1= tf.reshape(z, (-1, 2,2,1024))
        #x1= tf.layers.batch_normalization(x1, training=is_train)
        x1 = tf.maximum(alpha * x1, x1)
        
        #print (x1.shape)
        x2 = tf.layers.conv2d_transpose(x1, 512,5, strides=2, padding='valid', kernel_initializer=tf.contrib.layers.xavier_initializer())
        x2 = tf.layers.batch_normalization(x2, training=is_train)
        #x2 = tf.layers.dropout(x2, rate= .2)
        x2 = tf.maximum(alpha * x2, x2)
        #print(x2.shape)
        
        x3 = tf.layers.conv2d_transpose(x2, 256,5, strides=1, padding='same', kernel_initializer=tf.contrib.layers.xavier_initializer())
        x3 = tf.layers.batch_normalization(x3, training=is_train)
        x3 = tf.maximum(alpha * x3, x3)
        #print(x3.shape)
        x4 = tf.layers.conv2d_transpose(x3, 128,5, strides=2, padding='same', kernel_initializer=tf.contrib.layers.xavier_initializer())
        x4 = tf.layers.batch_normalization(x4, training=is_train)
        x4 = tf.maximum(alpha * x4, x4)
        #print(x4.shape)
        
        
        

        
        #print(out_channel_dim)
        logits = tf.layers.conv2d_transpose(x4, out_channel_dim, 5, strides=2, padding='same', kernel_initializer=tf.contrib.layers.xavier_initializer())
        # 28x28x
        
#         print(logits.shape, 'logits shape')
        output = tf.tanh(logits)
        
#         print(logits)
#         print(output)
    return output


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    # TODO: Implement Function
    smooth = 0.1
    g_output = generator(input_z, out_channel_dim, is_train=True)
    
    d_output_real, d_logit_real = discriminator(input_real, reuse= False)
    d_output_fake, d_logit_fake = discriminator(g_output, reuse= True)
    
    
    
    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_real, labels=tf.ones_like(d_output_real)*(1-smooth)))
    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_fake, labels=tf.zeros_like(d_output_fake)))
    
    g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_fake,labels = tf.ones_like(d_output_fake)))
    
    
    d_loss = d_loss_real + d_loss_fake
    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # TODO: Implement Function
    # Get weights and bias to update
    t_vars = tf.trainable_variables()
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    g_vars = [var for var in t_vars if var.name.startswith('generator')]

    # Optimize
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)

    return d_train_opt, g_train_opt
    


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

from time import gmtime, strftime
import matplotlib.pyplot as plt

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # TODO: Build Model
  

    
    #print(data_shape)
    step = 0
    real_input,z_input,lr = model_inputs(data_shape[1],data_shape[2],data_shape[3], z_dim)
    
    d_loss, g_loss = model_loss(real_input, z_input, data_shape[3])
    
    d_train_opt, g_train_opt = model_opt(d_loss, g_loss, learning_rate, beta1)
    
    
    d_loss_change =[]
    g_loss_change =[]
    
    d_loss_local_change =[]
    g_loss_local_change =[]
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            
            for batch_images in get_batches(batch_size):
                #print(batch_images.shape)
                
                
                step += 1
                # TODO: Train Model
                batch_z = np.random.normal(0,.5 , size=(batch_size, z_dim))

                #print('batch_z shape',batch_z.shape)
                batch_images = batch_images * 2
                # Run optimizers
                _ = sess.run(g_train_opt, feed_dict = {z_input: batch_z, real_input: batch_images, lr: learning_rate})
                _ = sess.run(g_train_opt, feed_dict = {z_input: batch_z, real_input: batch_images, lr: learning_rate})
                _ = sess.run(d_train_opt, feed_dict = {real_input: batch_images, z_input: batch_z,lr: learning_rate})
                
                d, g = sess.run([d_loss, g_loss], feed_dict={z_input: batch_z, real_input: batch_images})
                #print(d,g)
                d_loss_change.append(d)
                g_loss_change.append(g)
                
                d_loss_local_change.append(d)

                g_loss_local_change.append(g)
                
                if step % 90 == 0:
                    # At the end of each epoch, get the losses and print them out
                    train_loss_d = d_loss.eval({z_input: batch_z, real_input: batch_images})
                    train_loss_g = g_loss.eval({z_input: batch_z})

                    print("Epoch {}/{}...".format(epoch_i+1, epoch_count),
                          "Discriminator Loss: {:.4f}...".format(train_loss_d),
                          "Generator Loss: {:.4f}".format(train_loss_g))
               
                    show_generator_output(sess,96, z_input, data_shape[3], data_image_mode)
    
#                     Discriminator_loss, = plt.plot(d_loss_local_change, color='b', label='Discriminator loss')
#                     Genereator_loss, = plt.plot(g_loss_local_change, color='r', label='Generator loss')
#                     plt.legend(handles=[ Discriminator_loss, Genereator_loss])
#                     print("****Local Iteration loss comparison****")
#                     plt.show()
#                     d_loss_local_change.clear()
#                     g_loss_local_change.clear()

    Discriminator_loss, = plt.plot(d_loss_change, color='b', label='Discriminator loss')
    Genereator_loss, = plt.plot(g_loss_change, color='r', label='Generator loss')
    plt.legend(handles=[ Discriminator_loss, Genereator_loss])
    print("****Whole epochs loss comparison****")
    plt.show()

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size =128
z_dim = 128
learning_rate = 0.0001
beta1 = .5
alpha= .2

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 4

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)
    
print('finsihed')

Epoch 1/4... Discriminator Loss: 0.9312... Generator Loss: 2.7862

Epoch 1/4... Discriminator Loss: 0.6037... Generator Loss: 4.5394

Epoch 1/4... Discriminator Loss: 0.5650... Generator Loss: 3.7329

Epoch 1/4... Discriminator Loss: 0.7126... Generator Loss: 3.1886

Epoch 1/4... Discriminator Loss: 0.6533... Generator Loss: 2.7279

Epoch 2/4... Discriminator Loss: 0.4748... Generator Loss: 3.3226

Epoch 2/4... Discriminator Loss: 0.5253... Generator Loss: 2.4119

Epoch 2/4... Discriminator Loss: 0.6705... Generator Loss: 2.0845

Epoch 2/4... Discriminator Loss: 0.6146... Generator Loss: 2.6126

Epoch 2/4... Discriminator Loss: 0.7212... Generator Loss: 2.1042

batch_size =64
z_dim = 128
learning_rate = 0.0001
beta1 = .2
alpha= .2

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 4

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)
    
print('finsihed')

Epoch 1/4... Discriminator Loss: 0.9452... Generator Loss: 3.1355

Epoch 1/4... Discriminator Loss: 0.9914... Generator Loss: 5.8404

Epoch 1/4... Discriminator Loss: 0.8481... Generator Loss: 1.7334

Epoch 1/4... Discriminator Loss: 0.9946... Generator Loss: 2.1700

Epoch 1/4... Discriminator Loss: 1.1395... Generator Loss: 2.4855

Epoch 1/4... Discriminator Loss: 0.6871... Generator Loss: 1.9092

Epoch 1/4... Discriminator Loss: 0.9170... Generator Loss: 1.4482

Epoch 1/4... Discriminator Loss: 0.9847... Generator Loss: 2.0798

Epoch 1/4... Discriminator Loss: 0.8663... Generator Loss: 2.3626

Epoch 1/4... Discriminator Loss: 0.7423... Generator Loss: 1.8770

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 160
z_dim = 256
learning_rate = .0001
beta1 = 0.3


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.0922... Generator Loss: 2.9755

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 0.5302... Generator Loss: 3.0356

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 0.4108... Generator Loss: 3.7168

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 0.3756... Generator Loss: 3.8924

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 0.3707... Generator Loss: 3.9059

****Local Iteration loss comparison****

batch_size = 160
z_dim = 256
learning_rate = .0008
beta1 = 0.4


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.2843... Generator Loss: 1.1432

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 1.3780... Generator Loss: 0.8028

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 1.3009... Generator Loss: 0.9997

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 1.3358... Generator Loss: 0.8474

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 1.3415... Generator Loss: 0.8802

****Local Iteration loss comparison****

batch_size = 160
z_dim = 256
learning_rate = .0001
beta1 = 0.3


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.0234... Generator Loss: 4.1481

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 0.5696... Generator Loss: 2.9624

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 0.4314... Generator Loss: 3.3216

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 1.3107... Generator Loss: 1.0508

****Local Iteration loss comparison****

Epoch 1/3... Discriminator Loss: 1.1484... Generator Loss: 1.1372

****Local Iteration loss comparison****

batch_size = 160
z_dim = 128
learning_rate = .0001
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.3241... Generator Loss: 2.6868

Epoch 1/3... Discriminator Loss: 0.5405... Generator Loss: 3.8324

Epoch 1/3... Discriminator Loss: 0.4615... Generator Loss: 3.3319

Epoch 1/3... Discriminator Loss: 0.3993... Generator Loss: 3.5335

Epoch 1/3... Discriminator Loss: 0.3916... Generator Loss: 3.6965

Epoch 1/3... Discriminator Loss: 0.3522... Generator Loss: 4.5467

Epoch 1/3... Discriminator Loss: 0.3604... Generator Loss: 3.8814

Epoch 1/3... Discriminator Loss: 0.3449... Generator Loss: 4.4902

Epoch 1/3... Discriminator Loss: 0.3420... Generator Loss: 4.6456

Epoch 1/3... Discriminator Loss: 0.3386... Generator Loss: 5.1356

batch_size = 160
z_dim = 128
learning_rate = .0001
beta1 = 0.4


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.1691... Generator Loss: 4.4858

Epoch 1/3... Discriminator Loss: 0.5444... Generator Loss: 2.8330

Epoch 1/3... Discriminator Loss: 0.4067... Generator Loss: 3.2293

Epoch 1/3... Discriminator Loss: 0.3926... Generator Loss: 3.8622

Epoch 1/3... Discriminator Loss: 0.3685... Generator Loss: 4.1066

Epoch 1/3... Discriminator Loss: 0.3569... Generator Loss: 3.9561

Epoch 1/3... Discriminator Loss: 0.3531... Generator Loss: 4.2038

Epoch 1/3... Discriminator Loss: 0.3439... Generator Loss: 4.6901

Epoch 1/3... Discriminator Loss: 0.3389... Generator Loss: 4.7748

Epoch 1/3... Discriminator Loss: 0.3372... Generator Loss: 4.9222

batch_size = 160
z_dim = 128
learning_rate = .0001
beta1 = 0.3


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 0.9004... Generator Loss: 4.1311

Epoch 1/3... Discriminator Loss: 0.5392... Generator Loss: 3.2758

Epoch 1/3... Discriminator Loss: 0.4329... Generator Loss: 3.1806

Epoch 1/3... Discriminator Loss: 0.3848... Generator Loss: 3.7272

Epoch 1/3... Discriminator Loss: 0.3712... Generator Loss: 3.9080

Epoch 1/3... Discriminator Loss: 0.3503... Generator Loss: 4.4217

Epoch 1/3... Discriminator Loss: 0.3499... Generator Loss: 4.0766

Epoch 1/3... Discriminator Loss: 0.3424... Generator Loss: 4.8684

Epoch 1/3... Discriminator Loss: 0.3375... Generator Loss: 4.9983

Epoch 1/3... Discriminator Loss: 0.3444... Generator Loss: 4.4320

batch_size = 160
z_dim = 128
learning_rate = .0001
beta1 = 0.2


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 0.8273... Generator Loss: 3.3608

Epoch 1/3... Discriminator Loss: 0.4719... Generator Loss: 3.6908

Epoch 1/3... Discriminator Loss: 0.4195... Generator Loss: 4.2444

Epoch 1/3... Discriminator Loss: 0.3944... Generator Loss: 3.5901

Epoch 1/3... Discriminator Loss: 0.3734... Generator Loss: 3.8278

Epoch 1/3... Discriminator Loss: 0.3517... Generator Loss: 4.3275

Epoch 1/3... Discriminator Loss: 0.3386... Generator Loss: 4.8590

Epoch 1/3... Discriminator Loss: 0.3427... Generator Loss: 4.8062

Epoch 1/3... Discriminator Loss: 0.3370... Generator Loss: 5.1965

Epoch 1/3... Discriminator Loss: 0.3332... Generator Loss: 5.5009

batch_size = 160
z_dim = 128
learning_rate = .0001
beta1 = 0.7


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.4081... Generator Loss: 2.7701

Epoch 1/3... Discriminator Loss: 0.9676... Generator Loss: 2.5690

Epoch 1/3... Discriminator Loss: 0.4814... Generator Loss: 2.9825

Epoch 1/3... Discriminator Loss: 1.5396... Generator Loss: 0.9701

Epoch 1/3... Discriminator Loss: 1.2084... Generator Loss: 1.0507

Epoch 1/3... Discriminator Loss: 1.1772... Generator Loss: 1.1676

Epoch 1/3... Discriminator Loss: 0.9969... Generator Loss: 1.3217

Epoch 1/3... Discriminator Loss: 1.1662... Generator Loss: 1.2520

Epoch 1/3... Discriminator Loss: 1.5942... Generator Loss: 0.7836

Epoch 1/3... Discriminator Loss: 1.3237... Generator Loss: 0.8625

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # TODO: Build Model
  

    
    #print(data_shape)
    step = 0
    real_input,z_input,lr = model_inputs(data_shape[1],data_shape[2],data_shape[3], z_dim)
    
    d_loss, g_loss = model_loss(real_input, z_input, data_shape[3])
    
    d_train_opt, g_train_opt = model_opt(d_loss, g_loss, learning_rate, beta1)
    
    
    d_loss_change =[]
    g_loss_change =[]
    
    d_loss_local_change =[]
    g_loss_local_change =[]
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            
            for batch_images in get_batches(batch_size):
                #print(batch_images.shape)
                
                
                step += 1
                # TODO: Train Model
                batch_z = np.random.uniform(-1,1 , size=(batch_size, z_dim))

                #print('batch_z shape',batch_z.shape)
                batch_images = batch_images * 2
                # Run optimizers
                _ = sess.run(g_train_opt, feed_dict = {z_input: batch_z, real_input: batch_images, lr: learning_rate})
                _ = sess.run(g_train_opt, feed_dict = {z_input: batch_z, real_input: batch_images, lr: learning_rate})
                _ = sess.run(d_train_opt, feed_dict = {real_input: batch_images, z_input: batch_z,lr: learning_rate})
                
                d, g = sess.run([d_loss, g_loss], feed_dict={z_input: batch_z, real_input: batch_images})
                #print(d,g)
                d_loss_change.append(d)
                g_loss_change.append(g)
                
                d_loss_local_change.append(d)

                g_loss_local_change.append(g)
                
                if step % 90 == 0:
                    # At the end of each epoch, get the losses and print them out
                    train_loss_d = d_loss.eval({z_input: batch_z, real_input: batch_images})
                    train_loss_g = g_loss.eval({z_input: batch_z})

                    print("Epoch {}/{}...".format(epoch_i+1, epoch_count),
                          "Discriminator Loss: {:.4f}...".format(train_loss_d),
                          "Generator Loss: {:.4f}".format(train_loss_g))
               
                    show_generator_output(sess,96, z_input, data_shape[3], data_image_mode)
    
#                     Discriminator_loss, = plt.plot(d_loss_local_change, color='b', label='Discriminator loss')
#                     Genereator_loss, = plt.plot(g_loss_local_change, color='r', label='Generator loss')
#                     plt.legend(handles=[ Discriminator_loss, Genereator_loss])
#                     print("****Local Iteration loss comparison****")
#                     plt.show()
#                     d_loss_local_change.clear()
#                     g_loss_local_change.clear()

    Discriminator_loss, = plt.plot(d_loss_change, color='b', label='Discriminator loss')
    Genereator_loss, = plt.plot(g_loss_change, color='r', label='Generator loss')
    plt.legend(handles=[ Discriminator_loss, Genereator_loss])
    print("****Whole epochs loss comparison****")
    plt.show()

batch_size = 160
z_dim = 128
learning_rate = .0001
beta1 = 0.4


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.0692... Generator Loss: 3.8789

Epoch 1/3... Discriminator Loss: 0.5506... Generator Loss: 3.7154

Epoch 1/3... Discriminator Loss: 0.4745... Generator Loss: 3.0883

Epoch 1/3... Discriminator Loss: 0.4512... Generator Loss: 3.0524

Epoch 1/3... Discriminator Loss: 0.3697... Generator Loss: 3.8383

Epoch 1/3... Discriminator Loss: 0.3590... Generator Loss: 4.4306

Epoch 1/3... Discriminator Loss: 0.3465... Generator Loss: 4.3845

Epoch 1/3... Discriminator Loss: 0.3527... Generator Loss: 4.3012

Epoch 1/3... Discriminator Loss: 0.3720... Generator Loss: 3.6843

Epoch 1/3... Discriminator Loss: 0.3415... Generator Loss: 4.9305

batch_size = 160
z_dim = 256
learning_rate = .001
beta1 = 0.4


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.1015... Generator Loss: 1.8386

Epoch 1/3... Discriminator Loss: 1.2833... Generator Loss: 1.1397

Epoch 1/3... Discriminator Loss: 1.0662... Generator Loss: 1.0624

Epoch 1/3... Discriminator Loss: 0.9263... Generator Loss: 1.1985

Epoch 1/3... Discriminator Loss: 0.7725... Generator Loss: 2.0079

Epoch 1/3... Discriminator Loss: 0.5490... Generator Loss: 1.8379

Epoch 1/3... Discriminator Loss: 0.3817... Generator Loss: 3.3958

Epoch 1/3... Discriminator Loss: 0.5047... Generator Loss: 2.0284

Epoch 1/3... Discriminator Loss: 0.3697... Generator Loss: 5.3881

Epoch 1/3... Discriminator Loss: 0.5905... Generator Loss: 2.6895

batch_size = 160
z_dim = 256
learning_rate = .0001
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.3072... Generator Loss: 2.3465

Epoch 1/3... Discriminator Loss: 0.6743... Generator Loss: 3.0634

Epoch 1/3... Discriminator Loss: 0.4403... Generator Loss: 4.0978

Epoch 1/3... Discriminator Loss: 1.5092... Generator Loss: 0.9113

Epoch 1/3... Discriminator Loss: 1.3027... Generator Loss: 1.0108

Epoch 1/3... Discriminator Loss: 1.2594... Generator Loss: 1.0164

Epoch 1/3... Discriminator Loss: 1.2675... Generator Loss: 0.9361

Epoch 1/3... Discriminator Loss: 1.1036... Generator Loss: 1.0693

Epoch 1/3... Discriminator Loss: 1.3029... Generator Loss: 1.1249

Epoch 1/3... Discriminator Loss: 1.3660... Generator Loss: 0.8292

batch_size = 80
z_dim = 256
learning_rate = .0001
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.9508... Generator Loss: 2.8722

Epoch 1/3... Discriminator Loss: 0.8713... Generator Loss: 3.1193

Epoch 1/3... Discriminator Loss: 0.7587... Generator Loss: 2.6665

Epoch 1/3... Discriminator Loss: 0.7248... Generator Loss: 2.2869

Epoch 1/3... Discriminator Loss: 0.6709... Generator Loss: 2.6237

Epoch 1/3... Discriminator Loss: 0.6726... Generator Loss: 2.7693

Epoch 1/3... Discriminator Loss: 0.6558... Generator Loss: 2.4188

Epoch 1/3... Discriminator Loss: 0.6563... Generator Loss: 2.3076

Epoch 1/3... Discriminator Loss: 0.6622... Generator Loss: 2.5135

Epoch 1/3... Discriminator Loss: 0.6602... Generator Loss: 2.4607

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    # TODO: Implement Function
    smooth = 0.1
    g_output = generator(input_z, out_channel_dim, is_train=True)
    
    d_output_real, d_logit_real = discriminator(input_real, reuse= False)
    d_output_fake, d_logit_fake = discriminator(g_output, reuse= True)
    
    
    
    d_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_real, labels=tf.ones_like(d_output_real)*(1-smooth)))
    d_loss_fake = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_fake, labels=tf.ones_like(d_output_fake)*smooth))
    
    g_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits = d_logit_fake,labels = tf.ones_like(d_output_fake)))
    
    
    d_loss = d_loss_real + d_loss_fake
    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

batch_size = 80
z_dim = 256
learning_rate = .0001
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)
    
    
print('finished')

Epoch 1/3... Discriminator Loss: 1.1751... Generator Loss: 2.8885

Epoch 1/3... Discriminator Loss: 0.8225... Generator Loss: 2.6306

Epoch 1/3... Discriminator Loss: 0.6879... Generator Loss: 2.3275

Epoch 1/3... Discriminator Loss: 0.6684... Generator Loss: 2.3612

Epoch 1/3... Discriminator Loss: 0.6679... Generator Loss: 2.4848

Epoch 1/3... Discriminator Loss: 0.6721... Generator Loss: 2.7138

Epoch 1/3... Discriminator Loss: 0.6587... Generator Loss: 2.4426

Epoch 1/3... Discriminator Loss: 0.6605... Generator Loss: 2.1662

Epoch 1/3... Discriminator Loss: 0.6636... Generator Loss: 2.5039

Epoch 1/3... Discriminator Loss: 0.6552... Generator Loss: 2.3565

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.